Expert Visioning: Coding, Highlighting, Representing

Learning to See, Seeing to Learn

Introduction

A student hands in an AI-generated essay, quoting a book that doesn't exist written by a ghost writer. A surgical resident trusts an AI blatantly hallucinating a diagnosis of esophageal cancer due to coughing and troubles swallowing. A jury convicts based on AI-analyzed evidence that fabricates patterns in street noises. These aren't merely technical glitches—they represent a crisis of perception that Goodwin's (1994) framework of "professional vision" uniquely illuminated decades before AI barged in on our society. Professional vision—expert vision—according to Goodwin is built from sustained embodied participation in a domain with others who have already developed competence through participation in social discourse that shapes what other experts fully perceive but may be invisible to the untrained eye.

AI challenges or disrupts professional vision by mimicking expertise without the underlying sustained perceptual training requiring judgment or capabilities with expert tools put to use in real life. By creating convincing but hollow simulations of expertise, non-experts can be easily hoodwinked into accepting a false representation of a situation—a patient presents with coughing and troubles swallowing and therefore has cancer. AI systems develop their own form of "probable (or possible) expert vision" through training data, fundamentally different from human professional vision developed through embodied practice. AI failures aren't really technical problems unless we receive AI text uncritically; they’re simply wrong conclusions to be expected from a machine that predicts the next word. When the norm is uncritical acceptance of generated output, AI texts should evoke profound epistemological concerns about knowledge acquisition and expert application.

Charles Goodwin (1943-2018) began his work in part by extending Erving Goffman's dramaturgical theories of social interaction. For Goffman, individuals “perform” for an “audience,” sometimes “front stage” (public facing), sometimes “backstage” (private self). Academically, in the 1970s the times were afire with critical energy, pushing beyond structural-functionalist frameworks that had dominated for decades, leaning on the Frankfurt School’s critical theory from the 1930s, a robust theory of societal interrogation conceived as Hitler walked onto the front stage, a theory insisting that linguistics, sociology, any subject matter with a focus on people in social settings shouldn’t just describe but interrogate power structures shaping human misery. The idea was to expose how seemingly neutral systems—education, medicine, law, political bodies, universities, you name it—were hidden sites of domination, reproducing inequality by masquerading as objectivity incarnate.

Samuel Bowles and Herbert Gintis's Schooling in Capitalist America (1976) is a seminal critique of education systems in capitalist societies. These researches argued that the structure and norms of schools mirror those of the capitalist workplace. For example, the fragmentation of school subjects corresponds to the division of labor in production processes. Social relationships in schools develop traits which shape the private self to expect a certain economic position. Certain teacher-student relationships are considered normal that vary by the economic location of the student population, especially with regard to subordination to authority. The "hidden curriculum" refers to implicit lessons schools teach, such as obedience, punctuality, and acceptance of inequality. Of course, these lessons are not part of any teacher’s lesson plans; students are immersed from the opening to the closing bell and in this way are socialized to inhabit roles that sustain capitalist relations—college and career ready. Bowles and Gintis contended that education primarily serves the needs of capital accumulation. The meritocracy, school as a great social leveler, is a convenient lie.

Goodwin’s (1994) work on expert vision is significant today because it shows how expertise isn’t just about seeing but about framing and how doctors, lawyers, scientists, professors, teachers, economists, etc., are trained to interpret the world in ways that reinforce their frames. Individual experts partake in in-the-moment uses of collective expertise that sharpens their individual vision. In the best cases, experts apply expert vision using their very best expert judgment. In the worst cases, power motives distort their vision, and they selfishly proffer what they know to be false under the guise of authority. That’s why we have peer review and appellate courts, grade appeals, rescoring of standardized tests, imperfect as probably human social practices are.

Goodwin showed how interactional practices (like a doctor’s diagnosis or a courtroom’s rituals or a teacher’s final exam) are materially consequential. Today’s AI hallucinations feel like a grotesque amplification of a deep conspiracy to undermine expert vision. If a surgeon trusts a hallucinated cancer diagnosis, it’s not just a tech failure—it’s a collapse of the social safeguards Goodwin analyzed: the way professionals are responsible for challenging or corroborating and contextualizing knowledge not from the distance of an ivory tower, but from close reading of phenomenon using individual expertise and a commitment to professional ethics.

The 1970s taught us to question whose interests were served by so-called objective systems. Unfortunately for schooling, since the infestation of schools by NCLB and the Common Core movement, well, as somewhat of an expert, I pronounce critical theory dead on arrival. Now, with AI, we’re seeing what happens when those objective systems are literally, impossibly, invisibly synthetically unaccountable, available to anyone with a spare twenty bucks a month for a bot and a computer—no human biases to deconstruct, just opaque algorithms masquerading as truth. It’s a chilling twist on the notion of expert vision. Goodwin would’ve been fascinated (and horrified).

Goodwin’s (1994) contribution in the form of a theoretical model of “professional vision” revealed how specialized communities develop distinct ways of seeing reality. His analysis of archaeologists identifying ancient post holes in dirt and of lawyers manipulating visual evidence in the Rodney King trial demonstrated that perception itself is not a natural psychological process but a socially constructed activity dependent on coding schemes, highlighting practices, and graphic representations.

Goodwin’s analysis of the elements of expert vision offers a lens for understanding our current AI integration challenge. First, machines fundamentally lack professional vision. They cannot truly "see" anything as teachers, scientists, historians, or mathematicians see things in real time. Their outputs may convincingly mimic disciplinary knowledge, but they do not embody knowledge in an ontologically meaningful way. The AI that "analyzes" poetry cannot perceive literary devices as an English teacher does—it can only statistically model what humans have previously said about literary devices. The AI that “articulates” the causes of WWI cannot speak from the vantage point of an expert individual human who has visited important sites, read and written about focused questions, held seminars over many years. It can “articulate” what many human experts would probably say. Unfortunately, many of our high schools and universities also value what experts would “probably say”—how close can a student come?—often presented as a multiple choice option, above what any single student thinks at the moment.

For educators navigating this terrain, the mandate is as clear as a paradox can be: They and their students must develop a professional vision for recognizing the absence of professional vision in AI systems. The most valuable thing teachers bring to AI-integrated classrooms isn't technological fluency but their irreplaceable, unique, expert ability to see mathematics in numbers, scientific reasoning in data, historical significance in primary sources—and to teach students the difference between algorithmic mimicry and disciplinary vision.​​​​​​​​​​​​​​​​

At the University of Pennsylvania in the 1970s, Goodwin grasped the profound nature of Goffman’s insights into individuals’ performing for one another and presenting themselves strategically, maintaining "face" through carefully managed impressions. Students who are incentivized to perform for teachers are strategically maintaining academic face. While Goffman gave us conceptual tools for thinking about interaction as performance, Goodwin looked into the matter and found that social performances depend crucially on visual access, gaze coordination, and embodiment—face to face, a circumstance disrupted by COVID and now by AI. His insight illuminated how specialized expert communities develop distinct ways of seeing through embodied engagement in practices that transform participation—immersion in a professional community—into professional knowledge and expertise.

In this essay I hope to do justice to Goodwins’s model of professional vision as an analytical tool. I want to show how early reading teachers develop or fail to develop the expert vision required for the job. I want to take a risk and try to articulate the expert vision of a Reading Recovery teacher, a teacher who works one-on-one with young children who need more expert attention than they can get in a classroom. What do these teachers see when they work with a six-year-old child slow in picking up the skill? Then I want examine the expert vision of a teacher implementing Sigfried Englemann’s Direct Reading Instruction curriculum. I conclude with a discussion of how one might apply the fruits of this analysis to a model of expert pedagogical vision on integrated AI. But first, some background.

Goodwin’s Functional View of Professional Expert Knowledge

The core of Goodwin’s model he called ‘coding systems,’ which are specialized frameworks that experts use to locate aspects of a complex phenomenon in distinct categories constituting their field of expertise. These systems shape what and how professionals attend to and how they interpret reality itself. Think of the measurements a nurse records before the doctor enters from the wings. The blood pressure cuffs, the scales, the thermometers, and the seemingly informal chit chat with a nurse occur within an abstract coding frame, a frame which the attending physician shares.

A coding system works by providing standardized categories and terminology that members of a profession use to classify phenomena within their domain. For archaeologists, this might be a color classification system for soil samples. For police, it might be categories like "aggression" and "compliance" to evaluate suspect behavior. These coding systems serve multiple functions. First, they simplify perception by filtering the overwhelming complexity of real-world situations into manageable pieces. Goodwin describes how archaeologists classify soil into specific Munsell color codes, reducing infinite color variations to a finite set of named color values that can be recorded and compared.

Second, coding systems create shared professional knowledge by ensuring that different practitioners "see" phenomena in similar ways. Two archaeologists in different countries can meaningfully discuss a "10YR 5/6 yellowish brown" soil without having to physically examine the same sample. A school psychologist can interpret the findings from an observational classroom visit made by another school psychologist because they share an expert focus and understand how to separate significant behavior from irrelevant actions.

Third, these systems embed theory and prior knowledge directly into the tools of active perception. This point is crucial. The categories provided by a coding system reflect what the profession has already determined must be seen to make a judgement. An archaeologist's color chart encapsulates decades of field experience about which color distinctions matter for identifying human activity in soil. The expert on site shares knowledge of how and why such tasks are undertaken. The reading specialist who uses a phonemic awareness inventory to chart behavior matches the categories on the reporting form which the collective of experts knows to be worth attending to.

Coding systems shape perception in ways that go beyond any individual cognition. They are not templates to overlay and check off by simple casual observation. For example, taking inventory of different manufacturers of automobiles on a used car lot requires little expert judgment, if any. But physical tools, forms, charts, and protocols that structure how professionals interact and make decisions about more complex phenomena require collective expert vision. The essay rubric codifies qualities of essays; the expert makes the judgment. Voice? Tone, Style, Coherence? Among experts these qualities ought to mean the same thing in New Jersey as in Nevada, right? The Munsell color chart names colors for archaeologists, it requires expert judgement, and it shares the same meaning in Italy and in Ecuador. It physically transforms how archaeologists apprehend soil by providing viewing holes that juxtapose standardized colors with actual samples. It amplifies the distribution of expert cognitive power. But in the end, the expert on site makes the judgment.

The power of coding systems lies partly in what they omit. Once observations are transformed into professional categories, the uncertainties, disagreements, and contextual specifics of the original observation disappear from the record. The messy reality becomes a clean data point, and alternative ways of seeing become increasingly difficult to articulate or even imagine. Through these processes, coding systems actively construct the mental objects of knowledge around which professions organize their practices, theories, and identities.​

Three Key Practices of Professional Vision

According to Goodwin's framework, professional vision involves three primary practices that work together to shape how experts perceive their domain. I’ll start with elaboration on the basic tool.

Coding Schemes

As tools, coding schemes are instruments in systematic practices that translate reality into the categories that define a profession. When archaeologists classify soil colors using the Munsell chart, they're applying a coding scheme that turns the infinite variation of dirt colors into standardized values to be recorded and compared. Similarly, when police experts classify subtle body movements as "aggression," they're using a coding scheme that stabilizes ambiguous physical actions into categorical judgments that assume the intention of the prospective criminal and justify particular responses. Goodwin’s analysis of the Rodney King arrest video at the coding level in the 1994 piece is a compelling piece of evidence for his model. In this case, the behaviors in question could be coded as evidence of aggression endangering police officers or not. More about this in a moment.

Highlighting

Highlighting involves making specific phenomena stand out against a complex background. Irrelevant data have already been stripped away during coding, but the task of judging what is significant remains. When an archaeologist establishes lines around a post mold in the dirt based on variations in soil color, a clearly visible color differential suggests somebody put a post in the ground thousands of years ago. The lines establish a clear figure-ground against a literally amorphous ground. The outline highlights the object and becomes a persuasive tool that guides others' perception while reifying the object that the archaeologist concluded is visible in the soil's color patterns. What was a patch of ordinary soil now has a unique artifact visible and knowable, separated from its random bed of earth by human perception, made available for easy perception by other experts.

In the Rodney King trial, the defense coded the actions unfolding in time in the video, and then stopped trade motion. The defense then inscribed white outlines on video frames to make King's body stand out from the scene, directing attention to his movements while pushing the officers into the background, making a judgment of aggression more palatable for non-experts on the jury. As he lay on the ground under assault by police officers, the white lines reframed his struggles, depicting movements of his hips in stopped motion, suggesting his struggle was active resistance.

Producing and Articulating Graphic Representations

The third practice involves creating material representations—maps, diagrams, photographs—to objectify and communicate professional judgments using highlighted data within coded categories. When archaeologists transfer measurements from a soil profile of an old post hole to graph paper, their representation becomes the official record of their observations, stabilizing fleeting perceptions into durable artifacts. In the Rodney King courtroom, the defense's enlarged, cropped photographs of video frames transformed rapid, murky video images into static, analyzable objects that could be scrutinized at length. These representations don't record observations but actively shape them, allowing certain aspects to be highlighted while others fade into invisibility.

Together, these three practices—coding, highlighting, and representing—form the foundation of professional vision, enabling experts to transform complex perceptual fields into the specialized objects of knowledge that define their disciplines.​​​​​​​​​​​​​​​​

The Professional Vision of Reading Recovery Teachers

Reading Recovery is a model professional development program designed to build expert vision among early reading teachers. Unlike preparation to teach early reading which focuses on content knowledge of concepts like phonemes, morphemes, vowel patterns, phonic regularities, and the like, Reading Recovery sets out to teach teachers how to listen to, watch, and perceive reader behaviors immediately relevant to their reading strategies. In particular, these teachers observe struggling reads while they are actually reading texts using what they know about texts work. Linda Darling-Hammond (2017), internationally respected as an expert on professional development for teachers, judged Reading Recovery to be one of two premier professional development systems based on a set of stringent criteria, including evidence of success with students. Darling-Hammond described the development approach as follows:

“To prepare teachers to play this critical role, Reading Recovery provides intensive PD that incorporates all seven of the elements of effective PD. In groups of 8 to 12, teachers complete a yearlong graduate-level training course taught by a literacy coach. This sustained training involves model lesson observation, teacher demonstration of effective teaching techniques, and frequent collaborative discussion between participants. After the training course, faculty from the partnering university support teachers in their classrooms and facilitate program implementation within their area. Additional, ongoing PD for these teachers includes a minimum of six sessions with a Reading Recovery teacher leader and colleagues; opportunities for interaction and collaboration with school leaders and colleagues; and ongoing access to conferences (p.5)”

The following is speculative fiction based on my knowledge of Reading Recovery. Though I haven’t been trained in this method, I have seen trainings in action, read extensively about it, and talked with Reading Recovery teachers. I have a reading specialist credential. I taught at Illinois State University, which was a regional training center for the approach, and my wife is a Reading Recovery trained teacher. I have had considerable clinical experience teaching struggling readings one-on-one. Despite this lack in my training, I judge myself qualified to offer this scenario and take full responsibility for any errors in details. I would much appreciate feedback from bona fide RR teachers and will amend and correct this document accordingly.

In the small, brightly-lit corner of a first-grade classroom, Mr. Moore sits beside six-year-old Jayden at a kidney-shaped table. This is their daily Reading Recovery session—an intensive intervention for struggling readers. To an untrained observer, it might appear to be simple one-on-one reading practice, but beneath the surface, a sophisticated system of professional vision is at work.

Coding the Reading Process

Mr. Moore opens Jayden's reading folder, retrieving yesterday's running record—a coding form with tiny boxes and cryptic symbols that transforms his reading behavior into analyzable data. As Jayden begins reading a simple book about a lost puppy, Mr. Moore’s pencil hovers above his notation sheet.

When Jayden reads "The dog runned away" instead of "The dog ran away," the teacher makes a quick mark on a form. He doesn’t stop Jayden. He doesn't simply note "mistake"—he applies a precise coding scheme that distinguishes between different types of errors. This particular substitution gets coded as a grammatical error that preserves meaning, a type of error which is safely ignored in the moment.

"That word tricked you," Mr. Moore says after Jayden finishes the page. "Let's look at it again." The glass is two-thirds full. Jayden recognized that he needed a verb in the slot, and he recognized that the meaning of that verb was “run.” However, he didn’t attend to the visual aspects of the word, the vowel “a” became “u,” and Jayden affixed the -ed inflectional ending. From the perspective of a teacher concerned only with counting a mistake, this difference between the observed response and the expected response is actually bad news. Phonics isn’t the issue. Jayden is responding to syntactic cues (the word is a verb) and semantic cues (“to run”). What Jayden is missing is knowledge of the irregular forms of the verb “to run.” He overgeneralized the regular pattern in which verbs become past tense (adding -ed) and read the word as he likely uses it. So the issue is less phonics and more grammar. Mr. Moore now knows something about Jayden’s reading performance that can be helpful in understanding his mindset toward reading.

As Jayden reads aloud, the teacher codes Jayden’s problem-solving strategies: Does he rely on initial sounds? Letter patterns? Grammatical or semantic cues? Each approach has its own shorthand notation in the system. These codes don't just actively shape what the teacher attends to. By focusing on specific error patterns and problem-solving attempts, other dimensions of Jayden’s reading—his expression, interest in the story, or physical posture—become background features that go largely undocumented.

Highlighting What Matters

"I noticed something really powerful you did on this page," Mr. Moore says, pointing to a specific spot in the text. This deliberate highlighting isn't random—it directs Jayden's attention to particular strategic behaviors that match instructional goals. Today, he highlights Jayden’s successful self-correction where he first said "jumped" but then fixed it to "hopped." Jayden initially ignored visual cues (the letters) and responded to semantic and syntactic cues. Mr. Moore didn’t highlight the miscue. He highlighted the self-correction, a highly prized behavior indicating self-monitoring of all three cueing systems. He also noted that Jason relied on grammar and semantic cues, but in a completely different context. Here, there is no undeveloped grammatical knowledge that explains the miscue. “Hopped” and “jumped” are both regular verbs forming their past tense by adding -ed. The fact that Jayden self-corrected—he went back and reread the word—is a cause for celebration. Jayden is attending to all of the language resources available to him and monitoring all three cueing systems.

When they move to the writing portion of the lesson, the teacher watches Jayden form the word "house." As he hesitates after writing the first three letters, Mr. Moore draws a box on his practice paper, sectioning off "hou" from the remaining space for "se." This physical act of highlighting transforms the blank paper into a structured space that makes certain visual features of the word more salient than others. It reinforces the idea that readers—and writers—must attend to all of the letters in a word.

Later, as Jayden attempts to write "black," Mr. Moore pulls out a plastic frame divided into boxes. "Let's push the sounds of that word," he says, prompting Jayden to move a counter into each box as he articulates the separate phonemes: /b/ - /l/ - /a/ - /k/. This material tool highlights the phonological structure of the word, making abstract sound units concrete and manipulable—and visible. This activity strengthens Jayden’s ability to segment phonemes and perceive how they blend together to form words and it reinforces the link between reading and writing words.

Throughout the session, Mr. Moore produces representations of Jayden's reading behavior—not just the running record form with its coding symbols, but also a "cutting book" where he records words Jayden can build and break apart, and an observation survey that quantifies his progress according to specific metrics. These artifacts construct what counts as reading in the professional vision of Reading Recovery.

By the time their 30-minute session concludes, Mr. Moore has transformed Jayden's reading behaviors into a series of notations, classifications, and judgments that will guide tomorrow's instruction—a process of professional coding and highlighting largely invisible to those outside this specialized and often misunderstood community of practice.​​​​​​​​​​​​​​​​

Comparing Reading Recovery to DI

Reading Recovery and Sigfried Engelmann's Direct Instruction (DI) are structured interventions for struggling readers, but they operate from much different professional perspectives and coding systems. Englemann’s system codes one category of decoding, the visual cueing system, not three. DI privileges successful decoding performance; reading recovery privileges metacognitive behavior. Reading Recovery's approach to teaching struggling readers demands extraordinary levels of teacher expertise through its intricate system of professional coding and highlighting unlike Engelmann's Direct Instruction, which provides teachers with scripted lessons and predetermined sequences. Reading Recovery requires teachers to develop sophisticated observation skills and make complex instructional decisions in real time.

The Coding Systems: Observational Analysis vs. Scripted Sequences

In Direct Instruction (DI), Engelmann created a highly structured system where teaching is organized around a faultless communication principle. Teachers follow carefully scripted lessons with precise wording, systematic example sequences, and choral responses from students. The coding scheme is embedded in the curriculum itself, with the teacher primarily responsible for executing the program as designed. As one DI proponent explains, "even minor changes in teachers' wording can confuse students and slow their learning." This approach deliberately minimizes teacher decision-making to ensure consistent implementation—essentially restricting the teacher's professional vision to recognizing when to advance through predetermined sequences. By contrast, Reading Recovery requires teachers to develop a sophisticated diagnostic vision. During each lesson, the teacher must conduct and interpret "running records" of a child's reading behaviors—a complex notational system that codes specific types of errors, self-corrections, and problem-solving strategies. The teacher must simultaneously:

1. Record precisely what the child reads and how it differs from the text

2. Interpret the nature of each error (visual, meaning-based, grammatical)

3. Note strategic behaviors like monitoring, searching, and self-correcting

4. Make real-time decisions about when to intervene

This running record system represents a far more demanding coding scheme than following DI scripts. Reading Recovery teachers must learn to see through this professional lens, developing what Marie Clay called a "theory of literacy learning" to guide their interpretations.

Highlighting: Predetermined vs. Responsive

In DI, highlighting primarily occurs through the program's design—the teacher's attention is directed to specific features of the curriculum by the script itself, which determines what to emphasize and when. The system is engineered to reduce variability in implementation.

Reading Recovery teachers, however, must highlight different aspects of literacy processing for each individual child. They must notice subtle patterns in a child's behavior, highlight these patterns for the child's attention, and create teachable moments through strategic questioning and prompting. As the Reading Recovery Council explains, teachers must base "ongoing instructional decisions and the selection of future books on these observations."

Professional Training: Technical Execution vs. Diagnostic Expertise

Reading Recovery's year-long training program is intentionally rigorous and transformative. The training involves intensive graduate-level coursework, one-on-one teaching with continuous feedback, and participation in "behind-the-glass" sessions where colleagues analyze a teacher's instructional decisions. Direct Instruction training, while thorough, focuses more on technical execution—ensuring teachers can deliver the scripted lessons with fidelity. The system is designed to work with teachers at varying skill levels by providing them clear direction, reducing the need for diagnostic decisions. There is no need for developing expert vision.

The Key Difference: Diagnostic Decision-Making

The fundamental difference lies in how each approach distributes pedagogical responsibility between the program and the teacher. In Direct Instruction, the cognitive load of sequencing, example selection, and wording is handled primarily by the program designers. The teacher's job is implementation with fidelity, following precise scripts and sequences.

In Reading Recovery, the teacher must constantly engage in complex diagnostic decision-making. The programme is different for every child. The child's competencies are the starting point, and the program moves towards what the child is trying to do as a reader. This approach requires teachers to develop what might be called a "professional diagnostic vision" that can both see and interpret subtle patterns in literacy behavior, then respond with precise teaching moves tailored to the individual learner. The Reading Recovery teacher doesn't “implement” a program—they must develop expert vision, continually analyzing, highlighting, and responding to the unique processing patterns of each child. This model represents a more complex professional vision than executing even the most carefully designed Direct Instruction script.​​​​​​​​​​​​​​​​

Professional Vision in the Age of AI: A Goodwinian Perspective

Charles Goodwin's framework of professional vision offers a powerful lens for understanding how teachers must develop new ways of seeing, interpreting, and responding to classroom learning in the age of AI integration. Just as archaeologists must learn to see post holes in soil stains and physicians must recognize pathologies in medical images, educators need to cultivate specialized ways of perceiving AI-integrated learning that transform complex classroom interactions into meaningful pedagogical knowledge.

Coding Schemes for AI-Integrated Learning

Teachers need new coding schemes that systematically categorize student behaviors and interactions with AI tools. This section is sparse primarily because we don’t know what these schemes should look like. These coding systems could capture:

AI Engagement Patterns:

How students initiate interactions with AI tools—whether as creative collaborators, information seekers, or task delegators—could be coded. Just as Reading Recovery teachers code specific reading miscue patterns, teachers might code whether students use AI to extend their thinking or to bypass thinking.

Algorithmic Literacy Indicators:

Teachers must develop categories for recognizing when students demonstrate understanding of AI capabilities and limitations. Can students effectively prompt, critique, and evaluate AI outputs? Do they recognize hallucinations or biases?

Learning Process Shifts:

How does AI integration affect metacognitive processes, persistence, and meaning-making? These aspects first require theorizing in the context of AI and exploring and then design specialized coding. Moments when students transition between independent thinking and AI-assisted work need particular attention.

Collaborative AI Behaviors:

Social dynamics around AI tools—who controls access, how discoveries are shared, whether AI becomes a "social actor" in group work—constitute another critical domain for professional vision.

Unlike Direct Instruction's scripted approach where coding is predetermined in curriculum design, effective AI integration requires teachers to develop diagnostic coding systems similar to Reading Recovery's running records, but tailored to these new domains of observation.

Highlighting What Matters in AI-Integrated Learning

Beyond coding, teachers must develop practices for highlighting significant moments in AI-mediated learning:

Critical Junctures:

Just as Reading Recovery teachers highlight strategic behaviors when children solve reading problems, teachers must learn to identify and highlight moments when students make significant decisions about AI tool use, particularly when they choose to rely on their own thinking rather than defauling to AI.

Prompt Engineering Breakthroughs:

When students develop effective prompts that demonstrate sophisticated understanding of AI capabilities, these moments should be highlighted and shared with the class.

AI Evaluation Practices:

Teachers should highlight when students critically evaluate AI-generated content, especially when they identify limitations, biases, or errors in AI outputs. The practice of highlighting transforms the abstract goal of "teaching critical AI use" into concrete classroom moments that can be recognized, named, and celebrated—making the invisible work of developing AI literacy visible to the learning community.

Graphic Representations for Communication and Feedback

The third practice in Goodwin's framework involves creating material representations that objectify professional observations for various stakeholders.

AI Interaction Diagrams:

Teachers need visual tools that map students' AI engagement patterns over time, showing developmental progressions in their AI literacy.

Prompt Quality Rubrics:

Visual matrices that categorize student prompts along dimensions of specificity, creativity, and effectiveness can provide clear feedback structures.

AI-Human Collaboration Maps:

Visual representations showing the division of cognitive labor between student and AI can help students reflect on their own patterns of dependency or independence.

Dashboard Visualizations:

For administrators and parents, simplified visualizations of classroom AI integration that highlight learning outcomes rather than technical processes can make abstract AI literacy concrete.

Teaching students to develop "expert AI vision"

Teaching students to develop "expert AI vision" represents perhaps the most crucial educational intervention in our AI-infused future. Students who cultivate this specialized perceptual framework learn to see beyond surface outputs, recognizing patterns in AI-generated responses and developing mental models of underlying algorithmic processes.

This expert vision enables learners to notice their own prompt habits, becoming metacognitively aware of when they're using AI as a genuine thinking partner versus merely seeking shortcuts. As their expertise deepens, students read AI "between the lines," developing sensitivity to subtle indicators of limitations or biases that would remain invisible to novices.

Importantly, those with expert AI vision can visualize their entire learning ecology, creating sophisticated representations of their own knowledge-building processes that position AI as one component within a broader system of resources. The ultimate pedagogical goal becomes metacognitive independence—students applying their own professional vision to AI interactions, making intentional, theoretically-informed choices about when and how to integrate these tools into authentic intellectual work.

In the Meantime

Goodwin's framework of professional vision offers a powerful structure for understanding what teachers need to learn as AI becomes a permanent fixture in educational settings. The development of specialized coding schemes, highlighting practices, and visual representations creates not just a new pedagogical skill set, but a fundamentally transformed way of seeing classroom activity.

Just as Reading Recovery teachers must learn to see beyond surface reading errors to the underlying cognitive processes at work, teachers in the age of AI must develop specialized perceptual frameworks that transform complex AI-human interactions into meaningful patterns that guide instruction. This professional vision will distinguish teachers who merely accommodate AI in their classrooms from those who leverage it as a powerful tool for deeper learning.

The challenge ahead is not implementation but perceptual transformation—helping educators develop new ways of seeing that make visible the otherwise invisible dimensions of AI-integrated learning. By adapting Goodwin's framework, we can begin to articulate what this transformed professional vision might entail and how to cultivate it systematically through teacher education and professional development.​​​​​​​​​​​​​​​​

The integration of AI into education revives the critical lens that animated Goodwin's contemporaries in the 1970s, challenging us to interrogate these seemingly neutral systems as potential sites of hidden domination. Just as Bowles and Gintis revealed how traditional educational structures reproduce inequality by masquerading as objective meritocracies, today's AI systems present themselves as impartial tools while potentially embedding and amplifying existing power asymmetries.

The Frankfurt School's insistence that we not just describe but actively interrogate structures shaping human experience becomes urgently relevant when algorithms make consequential decisions without transparent accountability. Developing professional vision for AI isn't a technical challenge but a political and ethical imperative—teaching students to perceive how these systems operate means equipping them to recognize when AI reinforces existing hierarchies or presents statistical probability as objective truth.

By cultivating expert AI vision, educators and students engage in precisely the kind of critical practice that exposes how technological systems, like the educational institutions analyzed by critical theorists decades ago, aren't neutral conduits of knowledge but powerful mediators that require constant interrogation regarding whose interests they ultimately serve.